How $BIRTH and the demand leg are built, what each does and does not claim, how the birth index relates to federal statistics, and what the free and licensed tiers permit. If something here is unclear or wrong, tell us.
A monthly index of realized births in the United States. DMX captures provisional birth data from state vital-statistics systems and from the federal provisional series that NCHS compiles from the same state records, archives every version of every file, and assembles the counts on a documented, versioned methodology.
It publishes as two series. $BIRTH-P is a direct count of registered births in the panel. $BIRTH-N is a national estimate scaled from that count, published with a 90 percent band.
$BIRTH is the supply anchor of the Net Labor Capacity family. The forward labor supply series (S-5, S-10, S-20) derive from it by cohort accounting; the demand series and composites are staged behind those. See the demand leg.
Birth cohorts are an input to school enrollment and future workforce estimates. Migration, survival and labor-force participation also shape those outcomes, and $BIRTH does not model them.
Same underlying events, assembled to a different standard. $BIRTH does not estimate US births independently of the federal system. Where a state publishes its own timely file, DMX reads it at the state registry. For most jurisdictions today it reads the federal provisional series, which NCHS compiles from the same state records.
Timing. DMX prints on a fixed day, the tenth US business day of the second month after the reference month. Provisional state-by-month counts usually appear earlier in the CDC WONDER query tool, which CDC says updates near the end of each month, and DMX does not claim to publish before it. Each print states what WONDER carried at its cutoff, and names any state-month WONDER did not yet carry as missing, not zero. The NCHS quarterly provisional report measures selected birth indicators rather than state-month counts, so DMX does not use it as a lead-time comparator.
Governance. Federal statistical products are built for statistical purposes. They are not versioned, parameter-frozen, revision-logged products designed to be licensed and relied on contractually. $BIRTH is.
$BIRTH contains no forecast of future births. $BIRTH-P is a count of registered births. $BIRTH-N is a model-based estimate of the current national total, scaled from that count and published with a 90 percent band.
The forward series (S-5, S-10, S-20) apply fixed survival, participation and hours schedules to people already born and counted. The schedules are assumptions about the future. DMX holds them constant between rebalancings, publishes them, and reports how sensitive each result is to them.
The demand leg follows the same rule, with one addition, because one of its two series is a model. See Is any of the demand leg a forecast?
The people who will make up the labor force twenty years from now are already born. Pension funds, insurers and labor-intensive corporates commit to liabilities decades out, and the size of the future workforce is an input to those commitments. DMX aims to supply that input as a timely, versioned and licensable series built from registered births.
The near horizons matter sooner than the twenty-year framing suggests. Birth cohorts drive childcare and pediatric demand within a year or two, school enrollment within five, and household formation and entry-level labor supply within twenty. Real estate, healthcare, education and staffing exposures all sit somewhere on that curve.
Institutions that make long-horizon workforce decisions: pension funds and their consultants sizing liabilities, insurers and reinsurers pricing long-dated risk, sovereign and macro funds, real estate developers and operators, healthcare and education systems planning capacity, and labor-intensive corporates doing long-range workforce planning. They work today from demographic projections, and DMX adds a monthly series built from registered births. The raw feed is designed for quant and macro research desks.
It is not built for retail use. Researchers, journalists and other individuals may register for the free tier in their own name.
Framework research prints for September and October 2026 are published. The status page lists them with their delay and correction notices. The registered $BIRTH Public print has not opened and no launch date is fixed. Paid data delivery is closed under Gate A of the readiness statement. Date not yet fixed
The print day is fixed even though the launch date is not: the tenth US business day of the second month following the reference month, 09:00 New York time, with the calendar published twelve months forward. Each print states what the CDC WONDER query tool carried at its cutoff.
The national $BIRTH print is free with registration, published for each reference month once the federal estimate covering it has appeared.
Proposed paid tiers are $25,000 a year for the raw index feed, $50,000 a year for regional plus supply series, and custom pricing for the full Net Labor Capacity suite. Founding rates are available to the first cohort on Tier 01, in exchange for a multi-year term and reference rights.
Pricing is indicative for anchor-client conversations. Paid data delivery has not opened. Final tier terms publish with the first registered print. Indicative, pre-launch
Not under the free tier, ever, and not under any tier before the first settlement print. The free-tier license bars making an index value part of the operative terms of any contract or financial instrument, so that it sets or adjusts a payment, price or other outcome.
$BIRTH is not a regulated benchmark. It is designed to align with the IOSCO Principles for Financial Benchmarks (versioned methodology, frozen parameters, published revision policy, archived vintages), which is the right standard for a series institutions rely on. Designing to a standard is not the same as being an administered benchmark under a regulatory regime, and DMX does not claim to be one.
Settlement values are staged, not hypothetical. Each jurisdiction’s count is read as it stood at a fixed settlement timestamp and never restated. Values are offered only once twelve months of DMX’s own archived vintages exist to set that timestamp, at an annual rebalancing announced in advance. From there, licensed state-level series (Tier 02) are settlement-grade where the jurisdiction grants redistribution in writing. If you need a settlement-referenced product, that is a conversation rather than a click-through license.
Official state vital-statistics systems, the same registries that produce the records behind federal birth statistics. DMX reads a state's own file where the state publishes one, through state data portals, open-data catalogs and direct data feeds, and reads the NCHS federal provisional series for the rest. In the current panel most members are read from the federal series. Fifty-two jurisdictions are surveyed: the fifty states, the District of Columbia, and New York City, which is a separate vital-records jurisdiction from New York State.
Every file DMX ingests is archived with a cryptographic fingerprint, twice daily, and never overwritten. That is what makes any published value reproducible exactly as it was known on any given date, including values a state has since revised or removed.
Because a birth is registered days to weeks after it occurs. Provisional counts for recent months are systematically incomplete and rise toward their final values as registrations arrive. Every provisional birth series in the world has this property. It is a fact about vital registration, not a defect in a particular source.
What differs is the treatment. DMX measures that filing curve from its own archive of prior vintages, so it can see how each jurisdiction’s counts filled in historically, month by month, and it treats the curve as part of the methodology rather than a footnote. On the site’s California chart, trailing months render as open points on a dashed line for this reason: shown as published, not adjusted, and marked as still filling in.
Yes, and the answer differs for the two things DMX publishes.
Free-tier and research prints can revise. Where a source jurisdiction revises its own published counts, the index value can move with it. Revisions are recorded in the revision log that accompanies the print. Any citation must carry the print date, because that is what identifies the vintage you relied on.
Settlement values do not revise. The construction is vintage-stamped rather than finality-seeking: it reads each jurisdiction’s count for a reference month as it stood at a fixed timestamp, and later revisions never alter the settled value. At launch that timestamp is 180 days after month end, with longer lags for a few jurisdictions whose filing behavior warrants it, both interim until confirmed on a year of DMX’s own vintages. A manifest-error window applies, and the per-jurisdiction extract publishes with every settlement print, so a settled value is reproducible from published material.
Every instrument we could find that settles on a government statistic works this way: inflation-linked Treasuries, ISDA inflation definitions, CME Class III milk, event contracts on CPI and payrolls. All settle on first-published or fixed-timestamp values and disregard later revision.
Between monthly prints, licensed tiers receive a recomputed research value each US business day for the reference months not yet printed. It is labeled indicative, carries its as-of timestamp and 90 percent band, and is the same construction as the print with more of the month’s data in it.
It is not a print. It is never a settlement value, it never appears in the free tier, and its movements are not revisions: a reference month becomes revisable only once it has been printed. The nearest analogy is a real-time index level against an official close, or a Fed nowcast against the official release.
No, and the number depends on the channel. A jurisdiction-by-jurisdiction survey found roughly 72.8% of 2024 US births reachable from sources that are public today, rising to about 97.2% once sources available on request are included. The panel is assembled from what DMX can actually source, and coverage is a published property of the index rather than an assumption inside it.
Coverage is not static. Jurisdictions publish at different frequencies and on different schedules, and some publish nothing provisionally at all. The methodology paper states how the panel is defined, how partial coverage is handled, and what the resulting error budget looks like.
Reachable is not redistributable. Of US births, 34.0 percent across 15 jurisdictions carry an explicit written grant to redistribute; California, 11.1 percent, is pending a counsel reading of its publisher’s terms. The licensed state-level product is bounded by that number and grows as jurisdictions confirm. The national print is not so bounded, because a national aggregate is a derived statistic rather than any one state’s series.
Residence. $BIRTH counts births to residents of a panel jurisdiction wherever the delivery took place. Under occurrence tabulation a resident of a panel state who delivers elsewhere disappears from the count, so coverage becomes a fact about where hospitals are rather than about which residents are covered.
Some states publish their timely provisional data on occurrence basis only; California, the largest jurisdiction, is one. For settlement the read follows a fixed order published per jurisdiction: the state’s own residence-tabulated file; otherwise the federal residence-tabulated provisional series; and only as a disclosed last resort the state’s occurrence count times a frozen basis factor, flagged in the settlement extract.
For the research print, occurrence-only sources are converted using a basis factor that is the observed ratio of a jurisdiction’s residence-tabulated count to its occurrence-tabulated count for the same period, computed from its own published finals rather than modeled from individual records. The factors are frozen at rebalancing and published in the methodology paper’s appendix.
It is a real risk, and it is disclosed rather than assumed away. DMX depends on data published by third parties with their own schedules, formats and priorities, and states do withdraw products. One state in the survey had removed a published archive without redirects while its own data guide still cited it.
Two things limit the damage. Every vintage is archived at ingest, so history already collected survives the source disappearing. And the methodology defines how the index behaves when a jurisdiction goes dark, rather than leaving it to be improvised in the moment.
The methodology paper is available on request now and publishes with the free tier. It states the construction, the parameters, the revision policy, and the limitations, including where the index is biased and by roughly how much. It is one specification covering both legs: the birth index and the supply series, and the demand series with their inclusion rules and the format of the public event log.
DMX discloses its construction rules, parameter schedules, known limitations and revision history. A reader can identify which inputs produced a published value and which assumptions would change it.
The index is designed to align with the IOSCO Principles for Financial Benchmarks: versioned methodology, frozen parameters between rebalancings, a published revision policy, and an archive that makes any historical value reproducible. Those are the properties an auditor or an investment committee would test.
DMX is not currently a regulated benchmark administrator and does not claim to be. Where a specific certification or control attestation matters to your diligence process, ask.
Two controls a settlement counterparty asks about first are already running. DMX captures every source on three separately hosted sites under one administrator. A file that can enter a settlement value is admitted only when the paired reads on two of those sites captured the same bytes. Each site’s capture record is signed and externally time-stamped so it cannot be backdated. Both are specified in the methodology paper’s quality controls, and the per-jurisdiction settlement extract lets anyone check a settled value against them.
The other half of Net Labor Capacity. $BIRTH and the S-series measure the hours that will exist. The demand leg measures the hours the economy will call for. Both resolve to the full-time-equivalent hour, so the two form a ratio.
DMX measures labor in full-time-equivalent hours. Automation can change the hours a task requires while the position remains. Measuring hours lets DMX separate those changes from complete job displacement.
D(t, h) = E(t) × G(t, h) × (1 − A(t, h)). E is aggregate FTE-hours from the BLS Current Employment Statistics at a stated vintage. G is the trailing ten-year compound growth in hours, frozen at rebalancing. A is the automation displacement share, the leg’s one judgment-bearing input, which is why it publishes in two versions that are never combined. Births enter here too at short horizons, through a frozen schedule of hours demanded in childcare, early education and pediatric care.
The leg follows the supply launch and is delivered at Tier 03. The methodology paper, inclusion rules and event-log format are available on request and publish with the free tier. The series are planned as licensed products. Follows the supply launch
They answer different questions that most AI-and-labor numbers run together without saying which. D-R is what happened: displacement an employer has itself attributed to AI or automation, under published inclusion rules, converted to hours. D-X is what is structurally at risk: for each occupation, a task-level exposure share, times a measured adoption rate, times that occupation’s share of hours, summed across the economy.
They are never blended. Realized and potential displacement are different kinds of evidence, and an index that averages them cannot be audited, because nobody can say afterwards which part of a headline came from a filing and which from a model. The distance between them publishes as its own series, DRR.
D-X measures potential displacement. Automating a task does not necessarily remove the hours attached to it, so D-X can exceed the hours actually removed. Comparing D-X with D-R can be informative, and the comparison corrects neither series’ measurement error.
Only under rules written to be applied without judgment, because D-R feeds the settlement construction and every inclusion has to hold up when a counterparty with money at stake asks why.
Every candidate event, included or not, is logged in public with its source documents, tier, weighting, corroboration and decision. Moving the series takes a real separation, explicitly attributed in an accountable document, corroborated at payroll scale, seasoned, diluted across twelve months, and recorded where the market can read it.
The same way the birth data does: DMX captures the source documents itself and archives every vintage. For D-R the sources are state WARN (Worker Adjustment and Retraining Notification) filings and SEC EDGAR. DMX surveyed 51 jurisdictions, the fifty states and the District of Columbia. The corrected October 2026 research print monitored 17 of them, 40.55 percent of nonfarm employment. Three have no public source, and 31 have no capture built yet. DMX archives the notice documents because the attribution language lives in them. Current hours come from the BLS Current Employment Statistics.
Not every jurisdiction publishes WARN notices, and not every feed is reachable every month. Every print carries a source-coverage statement: what was monitored for the period, what was captured, what failed, and what is structurally unavailable. D-R is defined over the monitored universe and never grossed up beyond it, so a coverage change shows as a disclosed change in scope rather than an adjustment to the tally.
D-X is built from published material: a task-exposure rubric applied to the O*NET task file, re-rated annually by disclosed frontier-model raters under a published aggregation rule, paired with adoption measured from a worker survey, an observed-usage index and BLS employment weights, cross-checked against Census firm-level data. The rubric, the rule, the rater identities, the O*NET version and the occupation-level shares publish with every print. Re-rating happens only at rebalancing, so a change in the exposure term is always a disclosed parameter change rather than a silent drift.
A is the automation displacement share: a dimensionless input, included displaced hours divided by current hours. D is the demand series: an output, in hours. The suffixes apply at both layers, so A-R feeds D-R and A-X feeds D-X. Where the homepage writes D-R and D-X it is naming the series; where a formula writes A-R and A-X it is naming the shares inside them.
The Displacement Realization Ratio is defined over the shares. DRR(t) = A-R(t) / A-X(t): realized, employer-attributed displacement divided by exposure-implied displacement. It compares two measures with different sources and limitations.
A-X can exceed the hours actually removed, and A-R omits displacement that employers do not name. Read DRR as the distance between two documented constructions. It does not measure the share of jobs lost to AI, and it does not establish that automation caused a change in employment. A low reading can reflect reallocation, augmentation or attribution lag. DRR is a standalone research series with its own history and revision log, and it stays outside settlement.
No forecast enters D-R. It is a tally of separation events that have already happened, filed by the employers concerned, recorded in a public log.
D-X is a model and publishes as one. It estimates what is structurally at risk now, exposure times adoption times hours, not what will happen. It carries uncertainty bands, built by a method the paper fixes before D-X launches, it is labeled research-tier, and it never enters a settlement value. The rule is the supply leg’s rule: what is counted or mechanical can settle, and what is modeled publishes beside it, labeled, never inside it.
The trend term G is where a forecast would most naturally hide, so it is worth naming. G is the trailing ten-year compound growth rate in hours, frozen at rebalancing: arithmetic on realized hours, not an outlook. Alternatives that would have imported an external projection or an arbitrary constant were considered and rejected for that reason.
No. No causal identification of AI’s effect on employment is claimed anywhere in the family. D-R measures displacement that employers have attributed to AI under published rules. D-X is an exposure model. The composites are defined over those two published constructions, not over an unobservable true effect, which is why neither needs to be a causal estimate for the index to be well defined.
D-R carries a bias that runs in both directions, disclosed rather than netted away. Firms have an incentive to cite AI when cutting costs for unrelated reasons, and to avoid citing it when displacement is real but reputationally costly. The mitigations are corroboration against payroll-scale evidence, occupation-level cross-checks, and public logging of every inclusion and removal; the residual direction of bias is stated in the paper’s limitations section. At launch, with public statements excluded, D-R under-counts displacement that was announced but never filed.
If you need an estimate of what AI has done to employment, you need a causal study.
Only D-R, only inside the Net Labor Capacity composites, and not yet. D-X and DRR are research series throughout and never enter a settlement value. Settlement composites use D-R alone; research composites publish in both a D-R and a D-X version, with uncertainty bands on the latter.
The composites are staged behind the two legs they are built from, because their settlement construction inherits the full defensibility burden of the attribution rules: source class, language standard, corroboration, seasoning and the public event log. No representation is made that the composites are settlement-ready until the event system has produced twelve consecutive official prints each carrying a fully seasoned settlement value, one complete pass of the twelve-month aggregation window. The composites’ base date is set separately from $BIRTH’s, because they depend on a D-R history the supply product does not.
As with $BIRTH, nothing in the free tier may be used as a settlement, valuation or reference input.
The national $BIRTH print, published monthly on DMX’s print day for the reference month six months back, once the federal estimate covering it has appeared. That is the two national series, panel-observed and national-estimate, side by side, with the 90 percent band, the coverage shares, the methodology paper and a summary revision log. It is free to use for your own or your organization’s research, planning and decision-making, and free to cite publicly with attribution.
It does not include indicative updates between prints, state-level or sub-national series, the monthly historical back-series, API or programmatic access, bulk download, the forward labor supply series, the demand series, or the Net Labor Capacity composites. Those are licensed products.
Registration does two things. It makes the license terms binding, including attribution, no redistribution and no settlement use. Birth counts are facts and facts are not copyrightable, so DMX’s control over the free print rests on contract.
It also tells DMX who uses the index and for what, which guides which jurisdictions and series to build next.
Registration asks for an address at your organization’s own domain, including your own domain if you work for yourself. Independent researchers and journalists without a work email can still register.
Registration is not open yet. Until it opens, the sign-up form takes a waitlist, and DMX emails you once when registration opens.
Source: $BIRTH Index, DMX Analytics Inc. Print of [publication date].
The print date is required, not optional. Values are subject to revision, so a citation without a print date does not identify a specific number and is not compliant attribution. Where the medium allows, link to dmxanalytics.info.
Four restrictions, and they are the ones that matter commercially:
The full terms govern. This is a summary, not a substitute for them.
Each proposed paid tier supplies the following. Tier 01: the national number on the print day, about four months before the free print carries that month, with daily indicative updates between prints, the full historical back-series, the complete revision log and API delivery for Python, R and Snowflake. Tier 02: state-level series, settlement-grade where the jurisdiction grants redistribution in writing, and the forward labor supply series at five, ten and twenty year horizons. Tier 03: the demand series, DRR and the Net Labor Capacity composites as each launches. Redistribution rights are included where DMX holds them to grant. Paid data delivery has not opened.
That the state series carries a settlement value under the same rules as the national one: the count read as it stood at that jurisdiction’s published settlement lag, never restated, reproducible from the per-jurisdiction extract, and protected by the two-site capture check.
Two qualifications, both stated in the methodology paper. State series are delivered only for jurisdictions whose published terms grant redistribution in writing, which is 15 jurisdictions and 34.0 percent of US births at launch; a sixteenth, California, is pending a counsel reading of its publisher’s terms. And which object a contract may reference depends on the size of the jurisdiction. For jurisdictions registering fewer than 25,000 births a year it is the settled trailing twelve-month sum rather than a monthly value, because a single small month is too easy to move by the timing of a few filings. At or above that threshold a contract may reference the monthly settled value or the trailing twelve-month sum. Both objects are published for every jurisdiction and for the national series, so a contract that adjusts once a year reads one published number rather than adding twelve of them itself.
Not yet, and not under any tier that exists today. Naming a published value inside a contract is a separate permission from the right to receive the data, and it is licensed separately as reference rights. reference rights not yet offered
The use is an ordinary one. A multi-year staffing agreement, a childcare operator’s lease, a public-sector wage agreement or a construction contract with a labor adjustment names a published series in an escalation clause and moves a price by its movement. Cost-of-living clauses have worked this way for decades. Reference rights permit the parties to name a settled DMX series in such a clause.
What the arrangement is, and is not. DMX is the benchmark administrator and, where the contract names one, the calculation agent, which is the party that computes and publishes the value the clause reads. DMX is never a party to the contract and takes no side of what the clause moves. Only settled values may be referenced, either the monthly settled value or the settled trailing twelve-month sum, and an adjustment date is a print date rather than a calendar date, because the settled value for a reference month exists only from the print at which it settles.
No such license is offered until the readiness conditions of the DMX operations manual are met, published in a dated readiness statement, and reviewed by counsel. If this is the use you have in mind, say so now. Hearing which of your own contracts would carry the clause, and what they are indexed to today, is useful to DMX well before the license exists.
Under a licensed agreement, once paid data delivery opens. The agreement states the scope explicitly. DMX can only pass through redistribution rights it has confirmed it holds from the upstream jurisdiction, and those rights vary considerably: some states grant reuse in writing, some grant it conditioned on attribution, some restrict commercial use specifically, and at least one bars redistribution outright.
DMX maintains a per-jurisdiction register of those rights, evidenced to a quoted-source standard and re-verified on a schedule, and the redistribution scope in your agreement is drawn from it. If a downstream product depends on a specific state, ask before signing: the answer comes from a documented row in that register.
A 30-day evaluation license is planned for every paid tier once paid data delivery opens: full access, no redistribution, and it converts or expires at the end. 30 days; 30–90 under discussion
The free tier supports academic and journalistic use.
Cite with attribution including the print date, do not republish the series itself, and if you need something the free print does not include for a specific piece of work, ask.
If the question you came with is not here, ask it. Unanswered questions are how this page gets better.